Expert AI Insights | https://xitter.cf/t.co/JVbOKvqTQB | aptaintern
Cambridge, England
Joined July 2024
- Tweets26
- Following64
- Followers850
- Likes97
Apta AI wins 2 categories at ETH Oxford Hackathon with our tool called Portent
Last weekend our team members Shirom, Vyas, & Akash participated in ETH Oxford Hackathon in competition with some of the best companies and devs from across the world. Within 36 hours they built portent in the, an AI-driven tool that predicts the impact of real-world events on cryptocurrency markets before the markets react. This enables traders and investors to manage risk and optimize their portfolios with greater foresight. Today we found out that Apta AI placed 2nd in the [Main Track] AI category as well as won the [subcategory Flare] External Data Source/Cross-Chain.
Check out the winners as well as our tool at dorahacks.io/hackathon/eth-o…. Congratulations to all the winners and shoutout to all the participants.
Apta AI retweeted
Join us on 11/04 with @heroaisearch & @WCTNoah at 4 PM EST/8 PM UTC❗️
Appearances by,🎙
-@adajonuse
-@aptaai
Set your reminders for our upcoming space!🐳
#MobyMedia #HeroAIsearch #AI
xitter.cf/i/spaces/1rmGPoqWayVKN
Apta AI retweeted
We put @Aptaai and @ChatGPTapp to the test by asking a simple question: "Where can I buy sauce?" 🍲
🔍 Results:
🔹 ChatGPT: Lacks crypto-contextual understanding. Even when prompted, it only provided generic exchange options.
🔹 Apta: Gave precise exchanges where you can actually buy sauce tokens. 🎯
Check out the video to see the full comparison!
(1/4) July 26: Google DeepMind's AlphaProof and AlphaGeometry 2 have shown unprecedented mathematical reasoning ability for challenging math questions. The International Mathematical Olympiad (IMO) is one of the hardest and most prestigious mathematical competitions, where the brightest students from each country compete to solve challenging math questions. AlphaProof and AlphaGeometry 2 solved 4/6 problems, including the hardest question, which only five other candidates solved- missing gold by only two marks. (deepmind.google/discover/blo…)
(2/4) July 23: Meta launches Llama 3.1, including the 405B model. This model achieves SoTA performance for various standard benchmarks and surpasses GPT-4 in various tasks. This marks an exciting opportunity for the AI community, where for the first time in a while, the best model is completely open-sourced for researchers and companies alike. With the model containing more parameters than the number of neurons in the brain, will we soon run out of data to train bigger models or will the trend of ever larger mammoth models continue? (ai.meta.com/blog/meta-llama-…)
Apta AI retweeted
OpenAI has released GPT-4o Mini, a cheaper alternative to ChatGPT 3.5
💬Vatsal Raina, CTO of @Aptaai for Hero.io and a top 1% Cambridge graduate in Information Engineering, weighed in:
"This is an exciting development from @OpenAI as they have now released a cheaper option than ChatGPT 3.5. It’s particularly promising to see OpenAI pursue the direction of smaller language models, which will permit low-cost AI to be deployed in real-world products outside academia."
➡️ He further explained that Apta is actively looking into model distillation approaches to offer low-latency and cost-efficient AI solutions.
"Model distillation is possible with the collection of real data for supervised training of smaller LLMs that mimic larger LLMs," Vatsal added.